The best AI image generators 2023

In recent months, there has been a surge in the development of AI image generators. However, the majority of these tools are based on or derived from a few fundamental models, specifically Midjourney, DALL-E2, and Stable Diffusion. Below is our compilation of the best AI image generators currently available:

AI image generator	Rating	Cost
1. Midjourney	4.6/5.0	$0.026–$0.05/image
2. DALL·E2	4.5/5.0	~$0.03/image
3. StarryAI	4.4/5.0	$0.008–$0.06/image
4. Adobe Firefly	4.2/5.0	Free
5. Deep Dream Generator	4.2/5.0	~$0.03/image
6. Stable Diffusion	3.6/5.0	Free
7. Craiyon	3.1/5.0	Free or $5 to $24/month

What’s the best AI image generator?

Currently, the best AI image generator is Midjourney because it supports the widest variety of styles, aspect ratios, and dozens of prompts that let you tweak every aspect of image generation. DALL·E2 and Adobe Firefly are close seconds, but they are much more restrictive in terms of what you can create and also perform worse overall.

The prompts used to test AI image generators

Considering that AI image generators will replace stock photos, the prompts should focus on situations where stock photos are commonly used. Here are five revised prompts to help you evaluate the image quality produced by each generator:

1. Office Environment: “A modern, open-plan office with diverse employees working together at their desks, using laptops and tablets.” This prompt will test the generators’ ability to create realistic and professional workplace scenes that can be used in various business contexts.
2. Fitness and Wellness: “A group of people participating in an outdoor yoga class at a beautiful park during sunrise.” This prompt evaluates the AI’s capacity to generate images related to health, fitness, and well-being, capturing the essence of the activity and the environment.
3. Technology and Gadgets: “A close-up shot of a person using a smartwatch, with the screen displaying various app icons.” This prompt will help you compare the generators’ ability to create detailed and relevant images of technology and gadgets that are commonly used in marketing and promotional materials.
4. Food and Cuisine: “A beautifully plated dish of gourmet pasta with fresh basil leaves and parmesan cheese on a rustic wooden table.” This prompt challenges the AI to generate visually appealing images related to food and culinary topics, highlighting presentation and attention to detail.
5. Travel and Landmarks: “A stunning view of the Eiffel Tower in Paris during a vibrant sunset, with tourists taking pictures in the foreground.” This prompt assesses the generators’ ability to create attractive and realistic travel images, showcasing famous landmarks and capturing the atmosphere of the location.

By using these prompts focused on common stock photo categories, you can effectively compare AI image generators’ performance in producing images that are applicable to a wide range of professional and personal uses.

Evaluating AI image generators: 4 key factors

To help readers effectively assess AI image generators, we present a comprehensive explanation of six crucial factors we used during the evaluation process:

1. Quality (40%): We examined the realism of generated images, considering aspects like accurate colors, textures, lighting, and shadows. We also assessed the level of detail and resolution in intricate areas to ensure high-quality standards. Furthermore, we focused on the coherence of the images in terms of perspective, proportion, and continuity within the scene. Lastly, we evaluated how well the AI image generator interpreted the given prompt, ensuring the produced images were suitable and relevant for their intended purpose.
2. User experience (30%): We evaluated the overall experience of using the AI image generator, taking into account the simplicity of the interface, the ease of providing prompts, and the clarity of instructions. We also considered customizability and processing time, ensuring the AI met users’ needs in terms of speed and adaptability.
3. Licensing and usage rights (20%): We underscored the importance of understanding the legal aspects of using AI-generated images. We reviewed the licensing terms and usage rights associated with the images, as these could impact their applicability and potential legal implications.
4. Price (10%): We compared the affordability of each AI image generator, considering subscription fees, pay-per-use pricing, or additional costs for premium features. We helped users evaluate if the price was competitive and offered good value for the quality and range of services provided.

By taking into account these four essential factors, users can effectively compare different AI image generators and make informed decisions based on their specific needs and preferences.

1. Midjourney (4.6/5.0)

How we tested Midjourney: We submitted the default prompts through a Discord service without specifying any parameters.

Midjourney is the best AI image generator that excels at producing aesthetically pleasing, painterly images with complementary colors, artistic use of light and shadow, and satisfying composition. Known for its consistency, Midjourney delivers coherent styles and appearances, making it ideal for creating a series of complementary images. Though not exactly photorealistic, the generated images come close enough to serve various design applications. Midjourney also boasts unique features not found in other AI tools, such as ultra-sharp image generation and an understanding of photography jargon.

However, Midjourney does come with a few drawbacks. Its free trial version can be chaotic and overwhelming, with numerous messages and images appearing on the screen. Additionally, the platform does not offer in-app editing capabilities, requiring users to export their images to another editing software. While its consistent results are often seen as a plus, they can also limit customization and novelty. Midjourney also operates via a Discord-based interface, which can be a learning curve for those unfamiliar with Discord. Lastly, the free account offered limited functionality, necessitating a subscription for commercial use and added benefits.

In the end, Midjourney boasts one of the largest AI communities of 14.5 million Discord members who submit 20 to 40 jobs each second.

» More: Midjourney review

Aspect	Rating	Comment
Quality	5.0/5.0	Each output contains four image variations in a compound resolution of 838×838px (for 1:1 ratio). You can then regenerate for the same prompt, create 4 new variations based on one of the variations, or upscale one of them to a full resolution. Midjourney is currently the most consistently outputs high quality images.
User experience	4.0/5.0	Processing time is about 30 seconds per image. The worst part about Midjourney is that you submit prompts and receive images through a shared Discord, so everyone can see your work and it gets lost in the flood of other people's prompts and images.
Licensing & usage rights	4.5/5.0	With a paid account, you can use images in any way you want, even commercially. The free account allows you only personal use.
Price	4.5/5.0	Processing time is about 30 seconds per image. The worst part about Midjourney is that you submit prompts and receive images through a shared Discord so that everyone can see your work and it gets lost in the flood of other people's prompts and images.

2. DALL·E2 (4.5/5.0)

How we tested DALL·E2: We entered the exact prompt into the input and considered the first four images created.

DALL·E 2 is an advanced AI system designed to generate images from text descriptions, offering a plethora of features that enable image enhancement and editing, such as "outpainting," "inpainting," and diffusion. Capable of creating almost photorealistic images with four times better resolution than its predecessor, DALL·E, this powerful tool can also imitate various artistic styles and manipulate existing images to produce new compositions or variations.

DALL·E 2's technology is built on a 3.5-billion-parameter model trained on numerous image-caption pairs from the internet, learning the relationship between visual concepts and descriptive text. A separate 1.5-billion-parameter model is employed to enhance the resolution of its digital images. The AI uses a process called diffusion to generate images by progressively adding and modifying random dot patterns.

While DALL·E 2 boasts impressive capabilities, it is still an evolving technology, and its performance may not be on par with the latest version of OpenAI. Incorrect data labeling can lead to false results, and when faced with unfamiliar text, the generated images may be significantly different from the intended outcome.

Aspect	Rating	Comment
Quality	4.5/5.0	AI images created by DALL·E are often quite realistic, and you can easily distinguish what they depict. However, they are not as life-like as MIdjourney's images.
User experience	4.0/5.0	Processing time is about 20 seconds per image. Quite often, you need to restart generating because you will receive an error notifying you that the website is overloaded.
Licensing & usage rights	5.0/5.0	You own the images you create with DALL·E, including the right to reprint, sell, and merchandise—regardless of whether an image was generated through a free or paid credit.
Price	4.5/5.0	~0.03/image. You start with 50 free credits and get 15 free credits each month. Additional credits cost $15 for 115 credits. Each credit gives you 4 images for one prompt.

Best AI image generators based on DALL·E 2:

• Shutterstock AI Image Generator (free to test, paid to license)
• Image Creator from Microsoft Bing
• Jasper Art (paid)

3. StarryAI (4.4/5.0)

How we tested StarryAI: The default settings of 50 runtimes, 0 seed, and portrait (4:5) we chose. The 2 best of the 4 images generated are presented in this post, but all four were considered during the evaluation.

Starryai uses the latest AI methods to transform text prompts into works of art, making the process of AI art generation simple and intuitive. Users simply enter a text prompt, and the app's artificial intelligence transforms the words into art. The app offers a variety of models, styles, aspect ratios, and initial images to customize the creations. Users can generate up to five artworks for free daily and without watermarks, and they have full ownership of their creations. The app is available for free on iOS and Android, and there is also a web tool available.

Aspect	Rating	Comment
Quality	3.5/5.0	The quality of StarryAI images is average, while face rendering is very poor even with a higher runtime setting. The model understands and follows prompts well, but the images lack a certain level of realism.
User experience	5.0/5.0	Processing time is about 20 seconds per image. You can submit multiple tasks and they will be worked on simultaneously, significantly reducing your waiting time if you have lots of prompts.
Licensing & usage rights	5.0/5.0	You own the copyright to your creations. You can use them for commercial and non-commercial purposes.
Price	4.5/5.0	$0.008–$0.06/image. Subscriptions are available in sizes from 200 to 8,000 generations per month, costing $8.99–$63.99/mo. Also available with monthly and annual (20% off) options. You get 5 free credits per day.

4. Adobe Firefly (4.2/5.0)

Adobe Firefly, unveiled in March 2023, is a cutting-edge suite of generative AI art tools designed to seamlessly incorporate artificial intelligence into Adobe's comprehensive range of applications and services, specifically for generating media content. As an innovative, AI-driven creative platform, Firefly is currently in its Beta phase, offering a broad array of generative tools to assist artists, designers, and other creatives in unlocking their imagination and transforming their ideas into reality.

Consisting of multiple AI models tailored for a variety of use cases, Firefly's initial model emphasizes the generation of images, text effects, and vector recoloring. Users can effortlessly alter images by typing commands, while Firefly generates content brushes, variations on existing images, and even potentially manipulates photos and videos based on user input.

Adobe Firefly aims to revolutionize the way we create and engage with digital art and design by integrating state-of-the-art artificial intelligence into the creative process. The platform is constantly evolving, with numerous additional features under development that promise to enrich the creative experience further. Although currently in beta and without definitive pricing, Adobe intends to incorporate Firefly into its existing suite of products, radically reshaping the creative landscape for artists and designers.

» More: Adobe Firefly review

Aspect	Rating	Comment
Quality	4.5/5.0	Firefly understands prompts well but often lacks realism. It's quite inconsistent too, sometimes delivering very life-like results and other times delivering very poor quality. However, it can produce almost every style, from photos and art to vectors. Images are also very detailed and high resolution. It also comes with a text effects feature and more coming soon that no other image generator offers.
User experience	4.8/5.0	It has one of the shorter processing times of about 5 seconds. The website is also fairly easy to use, but due to the high number of available styles and other options, it takes some time to find your way around them and combine the best ones.
Licensing & usage rights	2.5/5.0	You cannot use images commercially yet and they come with a prominent watermark.
Price	5.0/5.0	Completely free for now.

5. Deep Dream Generator (4.2/5.0)

How we tested Deep Dream Generator: We used the "Text 2 Dream" tab and entered each prompt. We used the "PhotoReal" AI model, landscape aspect ratio, and high quality and left the negative prompt, face enhance, and upscale & enhance on default.

Aspect	Rating	Comment
Quality	4.0/5.0	By default, you get around 1MP images. You can choose to upscale them to 1MP (??), 2MP, or even 5MP, which is plenty for most uses. The quality is very good, but the images are still distinguishable from the real-life photos. The model understands the prompt well, following the instructions to generate a photo. However, it sometimes misses a part of instructions (in image 1, there are no people, although we asked for them). You can create a variety of styles, including artistic, photographs, and fantasy. You can also ask it to enhance images or create an image by showing it an example image you like.
User experience	4.0/5.0	Processing time is about 20 seconds per image. The website is easy to understand, but some functions are hidden in tabs that you won't find right away. The regeneration of "energy" credits prevents you from submitting lots of requests at once, so you have to wait for them to recharge.
Licensing & usage rights	4.5/5.0	Commercial purposes are allowed only with the paid plans. Read more
Price	4.5/5.0	~$0.03 per image. Available with subscriptions ($19–$99/mth) and one-time packs ($19–$79). Also available for free (around 5 images per day).

6. Stable Diffusion (3.6/5.0)

Stable Diffusion, released in 2022 by Stability AI, is a deep learning, text-to-image model. It generates detailed images based on text descriptions and can be applied to tasks like inpainting and outpainting. The model's architecture includes a variational autoencoder (VAE), U-Net, and an optional text encoder. The VAE compresses images to a latent space, Gaussian noise is added, and the U-Net denoises the output. The VAE decoder then generates the final image. Text prompts can guide image synthesis through diffusion-denoising.

Stable Diffusion is an open-source model that runs on gaming PCs with a GPU of at least 6GB VRAM. It produces results comparable to DALL-E 2 and MidJourney but lacks a polished user interface. Users have rights to generated images, although the model has stirred controversy over ownership ethics and the impact on human artists. Stable Diffusion is more permissive in content generation compared to other AI-based products.

Aspect	Rating	Comment
Quality	3.0/5.0	All Stable Diffusion images are in 768×768px resolution. All images are visibly AI-generated and lack realism. The model understands the prompt well, following the instructions to generate a photo. However, it sometimes misses a part of instructions (in image 1, there are no people, although we asked for them). You can create a variety of styles, including artistic, photographs, and fantasy. You can also ask it to enhance images or create an image by showing it an example image you like.
User experience	3.5/5.0	Processing time is about 70 seconds per image. The tool is easy to use and is a part of HuggingFace. However, it's often overloaded, so you have to submit the request several times before you get into a queue (which accepts ~100 generations).
Licensing & usage rights	4.0/5.0	Commercial purposes are allowed only with the paid plans. Read more
Price	5.0/5.0	Free.

Best AI image generators based on Stable Diffusion:

• DreamStudio (free and paid)

7. Craiyon (3.1/5.0)

Craiyon is an AI image generator that can draw images from any text prompt that you enter. It was developed to be a lighter version of OpenAI's DALL-E and is designed to be as easy to use as the original DALL-E. Craiyon was initially trained on millions of images from the internet and their accompanying captions. These captions lead the model to choose images to use based on the text prompts. The AI was fed a database of images and text descriptions until it learned to associate words with not only shapes but also color combinations, line thicknesses, differences in perspective, and other elements of artistic style.

Craiyon is a free-to-use tool for non-commercial purposes, with paid subscription tiers for those interested in commercial use cases. The Craiyon model is free to use for non-commercial purposes. The paid tiers offer lower wait times for image generation. The generated images will appear on the website, and users can share their results with the community on Discord.

Craiyon is still in its early stage of development and is not a professional instrument that can be used to solve critical problems. It is more like a toy, but it is a demonstration of what we can expect in the future.

Aspect	Rating	Comment
Quality	1.5/5.0	Craiyon's images are good enough so that you can guess what they represent, but they are far from realistic. Each generation produces 9 variations of 1024×1024px resolution.
User experience	3.5/5.0	Processing time is 60 seconds per image and 30 seconds with the paid plan. The website is straightforward, yet the generation takes longer with the free version.
Licensing & usage rights	5.0/5.0	You can use images commercially or on social media with free and paid plans.
Price	4.5/5.0	Free or $5 to $24 per month for faster generation, no watermark, high priority, and privacy.

Performance evaluation of AI image generators

AI image generators have revolutionized the field of computer graphics and image synthesis. These sophisticated algorithms are capable of producing incredibly realistic images based on text descriptions or other inputs. Despite their impressive capabilities, AI image generators have varying levels of performance when generating specific types of images. In this section, we will discuss the key takeaways of the AI image generators' performance with a focus on food, artistic scenes, faces, texts, and anorganic shapes.

1. Food and artistic scenes

AI image generators excel at creating visually appealing images of food and artistic scenes. The primary reason behind this is the inherently flexible nature of these subjects. Since there are fewer strict rules governing the appearance of food items and artistic compositions, minor inaccuracies or inconsistencies are less noticeable. The AI algorithms can also tap into a vast database of pre-existing images, allowing them to generate a wide variety of realistic and creative outputs.

2. Faces

AI image generators tend to struggle when it comes to generating images of human faces. This is because human faces have a complex structure, with many subtle details that need to be accurately captured to create a convincing representation. Additionally, humans are exceptionally good at recognizing faces, making it easy for us to spot even the smallest inconsistencies or inaccuracies in an AI-generated face. Consequently, AI-generated faces often appear unnaturally distorted or exhibit the uncanny valley effect.

3. Texts

While AI image generators can recognize individual letters and produce text-like images, they often fail to generate coherent or meaningful texts. This is primarily due to the fact that generating meaningful text requires an understanding of language and context, which is a separate domain of expertise from image generation. As a result, AI-generated texts may appear visually accurate but lack any discernible meaning or message.

4. Inorganic Shapes

AI image generators also face challenges when it comes to generating images of inorganic shapes, such as grids or geometric patterns. These shapes typically follow strict mathematical rules and require a high degree of precision to be accurately represented. AI image generators, which rely on pattern recognition and probabilistic approaches, can struggle to generate the exact shapes and alignments necessary for accurate anorganic shape representation. This often results in images with noticeable inaccuracies or inconsistencies.

In summary, AI image generators demonstrate varying levels of performance depending on the subject matter. They excel at generating images of food and artistic scenes due to the flexibility and creativity associated with these subjects. However, they face challenges when it comes to generating images of faces, texts, and inorganic shapes, which require a higher degree of precision and understanding. As AI technology continues to advance, it is likely that image generators will improve in these areas, further expanding their potential applications in various industries.

What are AI image generators?

AI image generators are tools that use artificial intelligence algorithms to generate images from text prompts. These generators are based on deep learning algorithms that have been trained on large datasets of images and their corresponding descriptions. The algorithms can receive input in the form of words, which they then process to generate an image. The entire process takes mere seconds, allowing users to see the results of their work immediately.

AI image generators can be used for various purposes, such as generating inspiration for digital marketers and content creators. They can also be used to create art, character art for tabletop games, or funky images for social media. AI image generators can increase productivity by allowing users to create stunning visuals in seconds without having to learn complex editing software. They can also save time and money while creating images that are appealing to a wide range of audiences.

There are many AI image generators available in the market, such as DALL-E 2, Midjourney, DreamStudio, and more. These generators use machine learning to create realistic photos and illustrations based on a set of text instructions. Some generators require only a few keystrokes to generate an image, while others include additional styles and parameters to their generators to make the results more unique. AI image generators have become incredibly popular over the past year because of their ability to create high-quality images in a matter of seconds.

Benefits of using AI image generators:

1. Time and cost efficiency: AI image generators can quickly produce high-quality images, reducing the need for hiring graphic designers or photographers and saving both time and money.
2. Ease of use: These tools are user-friendly and do not require extensive technical knowledge, allowing a wider range of individuals to create visually appealing images.
3. Customization: AI image generators can combine different styles, lighting, and color inputs, allowing users to create a specific result tailored to their needs.
4. Large-scale generation: AI image generators can produce a variety of images in a short amount of time, making them ideal for large projects or campaigns.
5. Unlimited creativity: These generators can create unique and original images that can capture the attention of audiences and spark their interest.

Drawbacks of using AI image generators

1. Job displacement: The rise of AI image generators may lead to job losses in the creative sector, as artists, photographers, and designers may be replaced by these automated tools.
2. Copyright issues: AI-generated images may not always be original, leading to potential copyright infringement and legal issues.
3. Quality and realism concerns: Generated images may not always look natural or realistic, which can limit their usefulness in certain applications.
4. Bias and discrimination: AI image generators are trained on pre-existing datasets, which may contain biases or discriminatory content. This can lead to biased or inappropriate image outputs.
5. Limited understanding: AI image generators may struggle with generating accurate images of humans or specific objects, as their contextual understanding is based on the data they were trained on. This can result in outputs that may not fully meet users' expectations.

In conclusion, AI image generators offer significant benefits, including time and cost savings, ease of use, and creative possibilities. However, it is crucial to be aware of the potential drawbacks, such as job displacement, copyright issues, and limitations in quality and understanding. Using AI image generators responsibly and ethically is essential to minimize these potential issues.

How do AI image generators work?

AI image generators use machine learning algorithms, specifically artificial neural networks, to generate new images based on input parameters or conditions. To train the AI image generator, a large dataset of images is used, which can include anything from paintings and photographs to 3D models and game assets. The dataset should be diverse and representative of the images that the AI image generator will generate. Once the AI image generator has been trained, it can generate new images based on a set of input parameters or conditions. These parameters can include things like style, color, texture, and shape. The input parameters can be set by a user or determined by the AI image generator itself.

To generate images, the machine uses two neural networks. The first neural network is used to create the image based on the text input by the user. The second neural network analyzes the generated image with reference images. By comparing the photos, the second neural network scores the generated image that the user sees. One method of image generation is Neural Style Transfer (NST).

NST uses a grouping of algorithms to recreate an existing image into a new style. For example, you use an input image (such as a photograph of Marilyn Monroe) and a style image (such as The Weeping Woman by Pablo Picasso). The output would be a new image that looks like Marilyn Monroe but is painted in the style of Picasso's painting.

What kind of input parameters are used to generate images?

AI image generators use a set of input parameters or conditions to generate new images based on the learned patterns and features from the training data. These parameters can include things like style, color, texture, and shape. The input parameters can be set by a user or determined by the AI image generator itself.

To generate images, the AI image generator uses a database of images to learn how to generate new ones. The program makes a process based on the images in the database and creates a new image. Most AI image generators allow users to type in a text prompt, and the model will generate images that match the description. Some models also allow users to change certain parameters like how many images they want to create, how many steps, and the size of the canvas. Overall, AI image generators use a combination of input parameters and learned patterns to create new and unique images.

Frequently asked questions